Research on the operation and obstacle avoidance of mowing robots based on intelligent algorithms is of great significance for improving the operating efficiency of intelligent mowing robots, reducing the manufacturing cost of intelligent mowing robots and liberating the labor force. However, the growth status of grass, such as height and color, changes constantly in its life cycle, making it difficult for the mowing robot based on single season and single scene to work and avoid obstacles to meet the actual needs. In addition, obstacles in the grassland have multiple types and scales, which also increase the difficulty of the mowing task. In response to these problems, this paper produced a full life cycle image data set of grassland through a self-made grassland data rapid collection platform, and then it can be used to train the built convolutional neural network model. We use the trained model for real-time identification of the environment during the operation of the mowing robot. The results show that the method not only meets the requirements of the intelligent mowing robot for accurate operation, but also achieves the effect of real-time obstacle avoidance. It solves the problem that the intelligent mowing robot needs to set the working boundary and needs to install a large number of expensive sensors, which broadens the practical engineering application range of the convolutional neural network intelligent algorithm.